Que4.15. Explain multilayer perceptron with its armature and characteristics. Answer Multilayer perceptron 1. The perceptrons which are arranged in layers are called multilayer perceptron. This model has three layers an input subcaste, affair subcaste and hidden subcaste. 2. For the perceptrons in the input subcaste, the direct transfer function used and for the perceptron in the retired subcaste and affair subcaste, the sigmoidal or squashed- S function is used. 3. The input signal propagates through the network in a forward direction. 4. On a subcaste by subcaste base, in the multilayer perceptron bias b( n) is treated as a synaptic weight driven by fixed input equal to 1. x( n) = ( 1, x1( n), x2( n),. xm( n)) T where n denotes the replication step in applying the algorithm. similarly, we define the weight vector as w( n) = ( b( n), w1( n), w2( n)., wm( n)) T 5. Consequently, the direct combiner affair is written in the compact form The algorithm for conforming the weight vector is stated as 1. still, is rightly classified into linearly If the utmost number of input set x( n). divisible classes, by the weight vector w( n)( that is affair is correct) also no adaptation of weights are done. w( n 1) = w( n) if wTx( n)> 0 and x( n) belongs to class G1. w( n 1) = w( n) if wTx( n) 0 and x( n) belongs to class G2. 2. else, the weight vector of the perceptron is streamlined in agreement with the rule. Architecture of multilayer perceptron 1.Fig.4.15.1 shows architectural graph of multilayer perceptron with two retired subcaste and an affair subcaste. 2. Signal flow through the network progresses in a forward direction, from the left to right and on a subcaste- by- subcaste base. 3. Two kinds of signals are linked in this network Functional signals Functional signal is an input signal and propagates forward and emerges at the affair end of the network as an affair signal. Error signals Error signal originates at an affair neuron and propagates backward through the network. 4. Multilayer perceptrons have been applied successfully to break some delicate and different problems by training them in a supervised manner with largely popular algorithm known as the error backpropagation algorithm. Characteristics of multilayer perceptron 1. In this model, each neuron in the network includes anon-linear activation function(non-linearity is smooth). utmost generally used non-linear function is defined by yj = 1 1 exp( MC) where vj is the convinced original field( i.e., the sum of all weights and bias) and y is the affair of neuronj. 2. The network contains hidden neurons that aren’t a part of input or affair of the network. retired subcaste of neurons enabled network to learn complex tasks. 3. The network exhibits a high degree of connectivity. Que4.16. How tuning parameters prompt the backpropagation neural network? Answer Effect of tuning parameters of the backpropagation neural network 1. instigation factor a. The instigation factor has a significant part in deciding the values of literacy rate that will produce rapid-fire literacy. b. It determines the size of change in weights or impulses. Machine Learning ways 4 – 17 L( CS/ IT- Sem- 5) , the smoothening is minimal and the If instigation factor is zero. entire weight adaptation comes from the recently calculated change. , new adaptation is ignored and former If instigation factor is one. one is repeated. Between 0 and 1 is a region where the weight adaptation is smoothened by an quantum commensurable to the instigation factor. f. The instigation factor effectively increases the speed of learning without leading to oscillations and pollutants out high frequence variations of the error face in the weight space. 2. Learning measure a. A formula to elect literacy measure is h = 2 2 2 1 2 .) m N N N Where N1 is the number of patterns of type 1 and m is the number of different pattern types. b. The small value of learning measure lower than0.2 produces slower but stable training. c. The largest value of learning measure i.e., lesser than0.5, the weights are changed drastically but this may beget optimum combination of weights to be overstepped performing in oscillations about the optimum. d. The optimum value of literacy rate is0.6 which produce fast literacy without leading to oscillations. 3. Sigmoidal gain , the input- affair relationship of If sigmoidal function is named. the neuron can be set as O = ( 1) 1 1 e) 4.16.1) where is a scaling factor known as sigmoidal gain. b. As the scaling factor increases, the input- affair specific of the analog neuron approaches that of the two state neuron or the activation function approaches the( Satisifiability) function. c. It also affects the backpropagation. To get canted affair, as the sigmoidal gain factor is increased, learning rate and instigation factor have to be dropped in order to help oscillations. 4. Threshold value in eq.(4.16.1) is called as threshold value or the bias or the noise factor. b. A neuron fires or generates an affair if the weighted sum of the input exceeds the threshold value. c. One system is to simply assign a small value to it and not to change it during training. d. The other system is to originally choose some arbitrary values and change them during training. Que34.17. bandy selection of colorful parameters in Backpropagation Neural Network( BPN). Answer Selection of colorful parameters in BPN 1. Number of retired bumps a. The guiding criterion is to elect the minimal bumps in the first and third subcaste, so that the memory demand for storing the weights can be kept minimum. b. The number of divisible regions in the input space M, is a function of the number of retired bumps H in BPN and H = M – 1. c. When the number of retired bumps is equal to the number of training patterns, the literacy could be fastest. d. In similar cases, BPN simply remembers training patterns losing all conception capabilities. Hence, as far as conception is concerned, the number of retired bumps should be small compared to the number of training patterns with help of Vapnik Chervonenkis dimension( VCdim) of probability proposition. f. We can estimate the selection of number of retired bumps for a given number of training patterns as number of weights which is equal to I1 * I2 I2 * I3, where I1 and I3 denote input and affair bumps and I2 denote retired bumps. Assume the training samples T to be lesser than VCdim. Now if we accept the rate 10 1 10 * T = 2 Which yields the value for I2. 2. instigation measure To reduce the training time we use the instigation factor because it enhances the training process. b. The influences of instigation on weight change is c. The instigation also overcomes the effect of original minima. d. The use of instigation term will carry a weight change process through one or original minima and get it into global minima. 3. Sigmoidal gain a. When the weights come large and force the neuron to operate in a region where sigmoidal function is veritably flat, a better system of managing with network palsy is to acclimate the sigmoidal gain. b. By dwindling this scaling factor, we effectively spread out sigmoidal function on wide range so that training proceeds briskly. 4. Original minima a. One of the most practical results involves the preface of a shock which changes all weights by specific or arbitrary quantities. , also the most practical result is to rerandomize the If this fails. weights and start the training each over. Que4.18. Write short note on unsupervised literacy. Answer 1. Unsupervised literacy is the training of machine using information that is neither classified nor labeled and allowing the algorithm to act on that information without guidance. 2. Then the task of machine is to group unsorted information according to parallels, patterns and differences without any previous training of data. 3. Unlike supervised literacy, no schoolteacher is handed that means no training will be given to the machine. 4. thus machine is confined to find the retired structure in unlabeled data by our- tone.